Alibaba DAMO Academy, in collaboration with institutions including Shengjing Hospital of China Medical University, has developed an AI model for liver cancer diagnosis named DAMO LiON (Liver DiagnOsis Network).
Clinical trials show that the model accurately identifies primary liver cancer and detects easily overlooked liver metastases. When deployed as a workflow-compatible diagnostic support tool, this AI can help reduce missed or delayed diagnoses and guide early clinical interventions. The breakthrough research has been published in the prestigious international journal Nature Medicine.
The liver is a common site for malignancies, including primary liver cancer and metastases originating from cancers in other organs, such as colorectum or pancreas. However, early diagnosis remains a challenge. Small lesions, compounded by cirrhosis, fatty liver disease, and complex anatomy, can cause even experienced radiologists to overlook critical indicators on contrast-enhanced CT scans.
To bridge this gap, DAMO Academy leveraged its years of expertise in medical imaging AI to develop DAMO LiON. Experimental results demonstrate that the AI model can sometimes achieve higher accuracy in identifying malignant tumors than radiologists alone in complicated cases.
During clinical testing, the AI-assisted system delivered major improvements, such as 27% faster image interpretation time for physicians and 11.5% increase in sensitivity for detecting malignancies. In addition, junior physicians aided by the system can achieve diagnostic performance comparable to that of senior specialists.
Notably, during a two-month real-world prospective clinical trial involving over 10,000 patients, LiON identified 15 small metastatic lesions that had previously been overlooked. Most of these lesions were small, with an average of 1 cm in size, and early detection allows the patients to receive timely, potentially life-saving treatments.
To achieve this level of accuracy, DAMO LiON employs an advanced network architecture. It captures global relationships between lesions and the entire liver while preserving fine local textures and boundaries. This dramatically improves performance in challenging cases involving fatty liver, cirrhosis, or complex anatomy. Furthermore, the AI iteratively fuses multi-phase imaging data to detect pixel-level differences across phases, precisely identifying tiny, transient lesions that appear only briefly during contrast-enhanced CT scans.
Since its establishment in 2017, DAMO Academy has pioneered the use of AI to detect subtle pathological features imperceptible to the human eye. The academy has successfully developed several oncology AI models, including DAMO PANDA for pancreatic cancer screening, DAMO GRAPE for gastric cancer screening, DAMO COCA for colorectal cancer screening. With DAMO LiON, the academy extends its medical AI capabilities beyond early screening and deep into the diagnostic domain, helping more patients receive timely, accurate, and effective care.
